A method is presented to extract contours of dendrites from images taken in situ during the growth of dendrites into supercooled melt. These contours will be used to compare models and simulations of crystal growth with real crystals growing in a three-dimensional melt. The method presented is useful for any application, where contours have to be extracted out of images taken from natural objects. The advantage of the presented method is that manual interactions are drastically reduced, no background corrections are needed and the position of the found contours is independent on the chosen extraction parameters. The procedure described in this paper takes six steps; (i) the images are preprocessed to enhance their contrast; (ii) all edge lines of the crystal are obtained by convolution with a Laplacian of a Gaussian succeeded by (iii) a zero crossing detection; (iv) the relevant edge line is extracted by flood filling the calculated lines; (v) this contour has to be cleaned and is recorded for future scientific investigations; (iv) in order to obtain the real proportions of the shape the contour has to be transformed to undo the distortions of the image taking process in the experiments. (C) 2003 Elsevier B.V. All rights reserved.